Use scatter/gather in the alignment step

[lbeltrame]

Description of feature

Splitting fastq files, mapping and then merging them (like bcbio used to do), can provide a significant speedup for alignment.
This needs to be post 1.0 as it's quite an invasive change.

The parts that need to be done:

• Run fastp unconditionally and split there
• Group splitted reads by sample (see sarek: https://github.com/nf-core/sarek/blob/b5b766d3b4ac89864f2fa07441cdc8844e70a79e/workflows/sarek/main.nf#L247)
• Align grouped reads
• Merge reads using GATK's MarkDuplicatesSpark (requires some tweaks to sorting, or it gets insanely slow)